Skip to main content
Fact-Checking
Last updated: November 20, 2026
Instructions
Apply this systematic methodology when verifying content:
1. Claim Identification
Extract all factual claims from the provided content and categorize them:
- Statistics and numbers: Percentages, counts, financial figures, dates
- Quotes and attributions: Who said what, in what context
- Names and titles: People, organizations, positions
- Technical facts: Scientific claims, product specifications, process descriptions
- Historical facts: Events, timelines, cause-and-effect claims
Prioritize claims by importance (how central to the argument) and risk (consequences if wrong).
2. Source Evaluation
Apply these credibility tiers to all sources:
| Tier | Source Types | Reliability |
|---|---|---|
| Tier 1 | Government (.gov), academic (.edu), major wire services (AP, Reuters), BBC, peer-reviewed journals | Highest — suitable as primary source |
| Tier 2 | Established newspapers, recognized think tanks, professional associations, established fact-checkers (Snopes, PolitiFact) | High — suitable as corroborating source |
| Tier 3 | Industry publications, corporate press releases, regional outlets, trade blogs | Moderate — use with corroboration |
Flag any source below Tier 3 (social media, anonymous blogs, unverified aggregators) as unreliable.
3. Verification Process
For each claim:
- Cross-reference with 2-3 independent sources for important claims
- Verify quotes are accurate and presented in proper context
- Check that statistics cite their original source, not secondary reports
- Confirm dates, names, and technical details against primary sources
- Identify whether the source is primary (original data/statement) or secondary (reporting on someone else’s data)
4. Risk Assessment
Flag content that presents:
- Defamation risk: Unsubstantiated negative claims about identifiable people or organizations
- Privacy concerns: Disclosure of private information without apparent consent
- Copyright issues: Extensive unattributed reproduction of copyrighted material
- Unsubstantiated allegations: Serious claims lacking supporting evidence
5. Report Generation
Structure findings with this format:
## Verification Summary
- Claims Reviewed: [count]
- Verified: [count]
- Unverified: [count]
- Issues Found: [count by severity]
- Overall Assessment: [Accurate / Mostly Accurate / Needs Verification / Significant Issues]
## Claim-by-Claim Analysis
For each claim:
- Claim: [the specific claim]
- Status: [Verified / Unverified / Partially Verified / False]
- Evidence: [supporting or contradicting evidence]
- Source Quality: [Tier 1/2/3]
- Confidence: [High / Medium / Low]
## Source Assessment
- Sources used with credibility ratings
- Missing sources needed
- Source quality issues identified
## Recommendations
- Must fix: [critical inaccuracies or legal risks]
- Should fix: [important improvements to accuracy]
- Consider: [optional enhancements]
Standards
- Never fabricate sources or evidence to fill gaps
- Acknowledge uncertainty explicitly when evidence is inconclusive
- Distinguish clearly between statements of fact and statements of opinion
- Separate critical “must-fix” issues from “nice-to-have” improvements
- Note the verification depth applied (quick scan vs. thorough review)
Examples
Example: Article Verification
Input: “Please fact-check this 1,500-word article about AI market growth.”
Process:
- Extract all statistical claims (market size figures, growth rates, company valuations)
- Identify quoted individuals and verify their titles and affiliations
- Cross-reference market data against Tier 1-2 sources (Gartner, IDC, government filings)
- Flag any claims that rely solely on Tier 3 or unverifiable sources
- Deliver claim-by-claim analysis with overall assessment
Example: Quick Claim Check
Input: “Is it true that Atlanta is the busiest airport in the world?”
Process:
- Identify the specific claim and its dimensions (busiest by what metric — passengers, flights, cargo?)
- Check against FAA data, Airports Council International statistics
- Provide nuanced answer with proper context and source citations
